EMS PM10 for Beijing
Many previous multi-model ensemble ozone and tracer studies proved the ensemble averaging technique performance compared to any single model forecast (Straume, 2001; Delle Monache et al., 2003; Mckeen et al., 2005; O'Neill et al., 2005). As a complementary approach in air quality forecastimprovement process, the present technical work tests the multi-model ensemble PM10 forecasts accuracy by comparing the ensemble single models (CMAQ, CAMx, NAQPMS) and ensemble average PM10 forecasts (ENS-AVE) to averaged hourly PM2.5 concentrations of four air quality observation sites in Beijing (Changping, Beiyi, Peking University and IAP tower). Although larges discrepancies are noted sometimesbetween simulated results and observed data, single ensemble models present good skill in simulating PM10 over Beijing. Forecasts on the two days (3rd and 17th August 2008, before and during the Olympics) where the largest discrepancies occur are analyzed through the statistical parameters below. This essentially aims at examining multi-model ensemble PM10 forecasting system performance on thesetwo critical days (before and during the Olympics). The following statistical measures are calculated for individual ensemble model and ensemble average PM10 concentrations based on hourly simulated results of concerned four observation sites in Beijing:
Normalized Mean Error (NME)
Unpaired Peak Prediction Error (UPPE)
where N is the number of hourlyconcentrations at a given observation station, Co(x,ti) and Cp(x,t) are respectively observed and predicted concentrations at the station located at x for hour ti . Co(x,t´)max and Cp(x,t´)max are respectively maximum 1-hr observed and predicted concentrations at the observation station over each considered time sequence over the day. NME is a rigorous analysis that matches predicted and observed PM10concentrations in both time and space, whereas the UPPE analysis compares the maximum observed and maximum predicted concentrations over daily time sequence at a single site. The averaged results of four observation stations in Beijing are averaged and presented on below figures.
The mean absolute error (MAE) of the wind direction is also calculated for individual meteorological forecasts (ensemblemodels) and for the ensemble average using measured wind directions from monitoring sites where increased PM10 concentrations are observed. The MAE of wind direction is defined as:
where N is the number of hourly concentrations at a given station, WDp(x,ti) and WDo(x,ti) are respectively the predicted and observed wind directions at the observation station located at x for hour ti. If thecalculated MAE is greater than 180 degrees (maximum error in wind direction), the absolute value of MAE minus 360 degrees is calculated for each hour and then the sum of the absolute errors is averaged over the total number of hours. This correction is necessary because the error in predicted versus observed wind direction can not be greater than 180. The MAE allows examining the influence ofmeteorological on the ensemble normalized mean error (NME). Note that daily results below are obtained by averaging calculated measures from individual observation sites in Beijing.
Results and discussion (in process)
Fig.2: Normalized Mean Error (NME)
Fig.3: UnpairedPeak Prediction Error (UPPE)
Fig.4: Mean Absolute Error (MAE) of the wind direction
Fig.5: Correlation between NME and MAE
A statistical analysis of the predicted PM10 concentrations and predicted wind directions was completed for both two days simulations....